Stream Generation

The pyETA GUI initiates a 22-channel Lab Streaming Layer (LSL) stream named 'tobii_gaze_fixation' for real-time eye-tracking analysis. This stream supports gaze tracking and fixation detection, with data sourced from either a Tobii eye tracker or a mock service.

---

Stream Generation

• Script: track.py

• Class: Tracker

  • In application.py, the start_stream method creates a StreamThread thread with user-defined parameters and starts it:

    self.stream_thread = StreamThread()
    self.stream_thread.set_variables(tracker_params=tracker_params)
    self.stream_thread.start()

  • StreamThread spawns a TrackerThread, which instantiates and runs a Tracker object from track.py.

• Stream Creation: When push_stream=True (set via GUI checkbox), Tracker creates an LSL stream or when --push_stream flag is passed in CLI pyeta track --push_stream

• Channel modification: Combines raw, filtered, and metadata into a 22-channel array.

  • Applies OneEuroFilter to raw gaze coordinates.

  • Computes velocity and fixation status using velocity_threshold.

  • Produces left_filtered_gaze_x, left_filtered_gaze_y, right_filtered_gaze_x and right_filtered_gaze_y used for fixation detection.

OneEuroFilter Algorithm

This algorithm reduces noise and jitters in raw gaze data while preserving responsiveness, which is later used for fixation detection.,

• Steps:

  1. Derivative: Calculates rate of change:

     current_derivative = (current_value - self.previous_value) / time_elapsed

  2. Derivative Smoothing: Applies fixed cutoff (1.0 Hz):

     alpha_derivative = self.smoothing_factor(time_elapsed, self.derivative_cutoff)
     filtered_derivative = self.exp_smoothing(alpha_derivative, current_derivative, self.previous_derivative)

  3. Adaptive Cutoff: Adjusts based on velocity:

     adaptive_cutoff = self.min_cutoff + self.beta * abs(filtered_derivative)

  4. Value Smoothing: Applies exponential smoothing:

     alpha = self.smoothing_factor(time_elapsed, adaptive_cutoff)
     filtered_value = self.exp_smoothing(alpha, current_value, self.previous_value)

---

Stream Properties

• LSL stream Name: 'tobii_gaze_fixation'

• Channels: 22

Channel Structure of the stream is described below:

	Channel Name	Type	Unit	Description
Left Eye
1	left_gaze_x	gaze	normalized	Raw X gaze position (0-1)
2	left_gaze_y	gaze	normalized	Raw Y gaze position (0-1)
3	left_pupil_diameter	pupil	mm	Pupil diameter
4	left_fixated	fixation	boolean	Fixation status (True/False)
5	left_velocity	velocity	px	Gaze velocity
6	left_fixation_timestamp	timestamp	s	Time of fixation start
7	left_fixation_elapsed	duration	s	Fixation duration
8	left_filtered_gaze_x	filtered_gaze	normalized	Smoothed X gaze position
9	left_filtered_gaze_y	filtered_gaze	normalized	Smoothed Y gaze position
Right Eye
10	right_gaze_x	gaze	normalized	Raw X gaze position (0-1)
11	right_gaze_y	gaze	normalized	Raw Y gaze position (0-1)
12	right_pupil_diameter	pupil	mm	Pupil diameter
13	right_fixated	fixation	boolean	Fixation status (True/False)
14	right_velocity	velocity	px	Gaze velocity
15	right_fixation_timestamp	timestamp	s	Time of fixation start
16	right_fixation_elapsed	duration	s	Fixation duration
17	right_filtered_gaze_x	filtered_gaze	normalized	Smoothed X gaze position
18	right_filtered_gaze_y	filtered_gaze	normalized	Smoothed Y gaze position
Screen Data
19	screen_width	screen	px	Screen width
20	screen_height	screen	px	Screen height
21	timestamp	timestamp	s	Data timestamp
22	local_clock	timestamp	s	Local system clock

---

Stream reading for plotting

• Script: reader.py

• Class: StreamThread

• Resolves and connects to 'tobii_gaze_fixation'

• Continuously pulls samples

• Upon requests, gaze data is being parsed.

# For gaze data
StreamThread.get_data()

# For fixation data
StreamThread.get_data(fixation=True)